An optimized pipeline for Topup geometric distortion correction
Poster No:
1684
Submission Type:
Abstract Submission
Authors:
Maeve McGowan1, Jack Grinband1
Institutions:
1Columbia University, New York, NY
First Author:
Co-Author:
Introduction:
Echo planar imaging (EPI) sequences suffer from image distortion due to B0 magnetic field inhomogeneities, susceptibility changes across tissue types, eddy currents, gradient coil nonlinearities, and cardiovascular pulsations[1,2], particularly in the phase-encode (PE) plane due to lower acquisition bandwidth per pixel. In fMRI, these distortions in brain geometry can lead to poor spatial normalization between functional and structural images, incorrect placement of regions of interest, incorrect tissue segmentation, incorrect localization of brain activity, and increased noise in group averages of brain activity.
While there are methods for correcting geometric distortion in real time during scan acquisition[3], most corrections are performed retrospectively. FSL's Topup tool, a widely-used method for unwarping distorted EPI data, estimates magnetic susceptibility by computing the field that maximizes similarity between images acquired in opposite PE-directions[4,5] and uses it to unwarp distorted data.
However, Topup relies on good linear registration between the opposite PE-direction images to estimate the distortion accurately. Importantly, greater distortion means greater potential for misregistration between the two opposite PE-direction images and subsequent errors in the Topup-estimated B0 field. Further, in practice, the field map estimate is noisy, resulting in less than perfect unwarping in the PE plane[6,7]. Finally, while Topup can significantly reduce distortion in the PE plane, it has no effect in the two orthogonal planes. We propose a new pipeline (Downunder) that reduces these three sources of error by applying standard distortion correction tools recursively, using outputs of each loop as reference information for the subsequent loop, implementing Topup unwarping in the slice encode direction, and applying nonlinear registration to the structural T2w image to account for any remaining distortions.
While there are methods for correcting geometric distortion in real time during scan acquisition[3], most corrections are performed retrospectively. FSL's Topup tool, a widely-used method for unwarping distorted EPI data, estimates magnetic susceptibility by computing the field that maximizes similarity between images acquired in opposite PE-directions[4,5] and uses it to unwarp distorted data.
However, Topup relies on good linear registration between the opposite PE-direction images to estimate the distortion accurately. Importantly, greater distortion means greater potential for misregistration between the two opposite PE-direction images and subsequent errors in the Topup-estimated B0 field. Further, in practice, the field map estimate is noisy, resulting in less than perfect unwarping in the PE plane[6,7]. Finally, while Topup can significantly reduce distortion in the PE plane, it has no effect in the two orthogonal planes. We propose a new pipeline (Downunder) that reduces these three sources of error by applying standard distortion correction tools recursively, using outputs of each loop as reference information for the subsequent loop, implementing Topup unwarping in the slice encode direction, and applying nonlinear registration to the structural T2w image to account for any remaining distortions.
Methods:
The first round of Topup correction used one of the SE images as the reference for registration, identical to standard Topup preprocessing. However, in subsequent rounds, a reference mask of the undistorted image was used for registration. The mask was computed by taking the average of the undistorted opposite-PE images from the prior iteration. The mask was then brain extracted, and binarized, and thresholded, using only voxels with displacements less than 80% of maximum displacement for calculating the cost function for registration (reference weighting: weighting = 1 for displacements < 80%; weighting = 0 for displacements ≥ 80%). This approach further minimized the effect of PE distortion on the cost function. After iterative Topup correction, the B0 field estimate from Topup was used to create a shift map in the slice-direction using the slice-encode bandwidth. The undistorted SE image was corrected in the slice-direction and subsequently linearly (FLIRT) and non-linearly (FNIRT) co-registered to the T2w image.
Results:
The resulting Downunder-corrected data (healthy subjects, n = 27) showed increased geometric similarity to the T2w structural image. The mean correlation ratio between distortion-corrected SE-AP and T2w images across subjects decreased with each iterative Topup application and the addition of slice-encode correction (Fig 1A), indicating improved structural similarity to the T2w image. The addition of FNIRT as a final step produced lower correlation ratio values than the respective Topup-only images (data not shown). To determine how much Downunder improves the geometric distortion over Topup alone, a 3mm isotropic ROI seed was placed in MNI space and the geometric errors were simulated for each subject. Fig 1B shows the effect of distortion and blur of the ROI across subjects with a peak percent overlap of only 31% and a blur diameter of about 12mm.
Conclusions:
Our results suggest that Downunder significantly reduces errors in structural geometry, registration, and localization of brain activity.
Modeling and Analysis Methods:
Exploratory Modeling and Artifact Removal 1
Motion Correction and Preprocessing 2
Keywords:
Statistical Methods
1|2Indicates the priority used for review
Provide references using author date format
[2] Embleton, K.V., Haroon, H.A., Morris, D.M., Ralph, M.A., Parker, G.J., 2010. Distortion correction for diffusion-weighted MRI tractography and fMRI in the temporal lobes. Hum Brain Mapp 31, 1570-1587.
[6] Hutton, C., Bork, A., Josephs, O., Deichmann, R., Ashburner, J., Turner, R., 2002. Image distortion correction in fMRI: A quantitative evaluation. Neuroimage 16, 217-240.
[1] Jezzard, P., Clare, S., 1999. Sources of distortion in functional MRI data. Human Brain Mapping 8, 80-85.
[3] Juchem, C., Nixon, T.W., McIntyre, S., Boer, V.O., Rothman, D.L., de Graaf, R.A., 2011. Dynamic multi-coil shimming of the human brain at 7 T. J Magn Reson 212, 280-288.
[5] Smith, S.M., Jenkinson, M., Woolrich, M.W., Beckmann, C.F., Behrens, T.E., Johansen-Berg, H., Bannister, P.R., De Luca, M., Drobnjak, I., Flitney, D.E., Niazy, R.K., Saunders, J., Vickers, J., Zhang, Y., De Stefano, N., Brady, J.M., Matthews, P.M., 2004. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage 23 Suppl 1, S208-219.
[7] Wu, M., Chang, L.C., Walker, L., Lemaitre, H., Barnett, A.S., Marenco, S., Pierpaoli, C., 2008. Comparison of EPI distortion correction methods in diffusion tensor MRI using a novel framework. Med Image Comput Comput Assist Interv 11, 321-329.